RESEARCH MEMO · AI Software / Enterprise AI
Microsoft Turns Agent Work into a Metered Service as Enterprise AI Adds Usage Billing to Seats
Microsoft Copilot × Usage-Based Billing × Managed Runtime | AI Software Research Memo | 29 September 2026
Public company covered: Microsoft Corporation (NASDAQ: MSFT) | Research focus: Copilot, Agent 365 and enterprise AI platform economics
Microsoft is not replacing subscriptions with consumption. It is adding Copilot Credits on top of the fixed-price USL: Cowork, Code, Autopilot and frontier models consume credits, while Managed Runtime supplies a governed application environment. The product architecture strongly validates a seats-plus-usage model; revenue, margin and free cash flow remain unproven.
Fixed-price everyday AI plus Copilot Credits for advanced agentic work
Cowork identifies Models, Runtime, Context and Tools
Managed Runtime is available in preview, without standalone pricing or revenue disclosure
01 · RESEARCH BRIEF
The one-minute brief
On 25 September, Microsoft separated everyday AI from advanced agentic work through two billing layers. The user subscription license (USL) covers fixed-price productivity capabilities, while Cowork, Code, Autopilot and frontier models such as Astra and Fable use Usage-Based Billing and consume Copilot Credits. Copilot Managed Runtime entered public preview, supplying Microsoft-hosted execution, identity, governed data access and lifecycle controls for apps and code created through Cowork, Code, Copilot Studio and compatible tools. FinOps controls connect spending policies, credit requests, model availability and usage history to the operating model. ACIS therefore marks the thesis that enterprise AI will expand beyond user seats into metered work as Strengthened. This is product and billing architecture evidence—not evidence that incremental revenue, gross margin or free cash flow has already materialized.
Audio transcript
The core of this memo is not that Microsoft has canceled seat pricing. It is adding metered advanced work on top of the seat. Everyday Copilot capabilities remain covered by the user subscription license. Cowork, Code, Autopilot and frontier models consume Copilot Credits. Managed Runtime places apps and code created through Cowork, Code and Copilot Studio inside the Microsoft 365 tenant boundary, where IT can govern identity, data connections, versions and policy. FinOps tools connect spending policies, credit requests, model permissions and usage history to the budget. This materially validates a move from a seat-only economy to a seats-plus-usage model, but it is not financial proof. Cowork is generally available, Code is still rolling through Frontier, Autopilot is in private preview and Managed Runtime is in public preview. The next evidence must connect production use and credit consumption to net new revenue, incremental gross margin and free cash flow after capital spending.
Known facts and open questions
- Confirmed
- Microsoft announced the new Copilot, Usage-Based Billing and FinOps for AI on 25 September 2026. September 29 is the memo date, not the event date.
- Billing boundary
- UBB complements rather than replaces the USL. Advanced experiences consume Copilot Credits and require a USL for access. Pay-as-you-go is listed at $0.01 per Copilot Credit, while task consumption varies with model, runtime, context and tool use.
- Maturity boundary
- Cowork is generally available; Code is rolling through Frontier; Autopilot is in private preview; Copilot Managed Runtime is in public preview. They should not be described as one fully deployed production stack.
- Not yet proven
- Microsoft has not disclosed full credit formulas, standalone pricing, customer adoption, incremental revenue, gross margin or free-cash-flow contribution for Code, Autopilot or Managed Runtime.
01 | Access and context
The USL provides fixed-price everyday AI access and connects identity, data and work context across Microsoft 365.
02 | Usage and runtime
Copilot Credits meter advanced agentic consumption; Managed Runtime provides a governed environment for apps and code.
03 | Governance and economics
Spending policies, credit requests, model permissions and usage history connect AI work to budgets, approval and value measurement.
| Layer | Official evidence | ACIS interpretation | Still unproven |
|---|---|---|---|
| Fixed access | The USL covers everyday AI and is required to access UBB experiences | Seats remain the entry point to usage revenue | Seat retention, pricing and net revenue after usage is added |
| Metered work | Cowork, Code, Autopilot and frontier models consume Copilot Credits | The value unit is expanding from users to sustained work | Actual credit usage, repeat work and unit margin |
| Managed runtime | Managed Runtime supplies hosting, identity and governance inside the tenant boundary | Runtime is becoming part of platform control | Standalone pricing, adoption scale and revenue attribution |
| FinOps control | Admins can govern spend, access, credit requests and consumption | Cost governance is becoming a prerequisite for production adoption | Comparable savings, business outcomes and return on spend |
02 · THESIS → EVIDENCE → UPDATE
What changed in the thesis?
From a seat-only economy to a seats-plus-usage model
- Prior thesis
- Enterprise AI monetization would not stop at one Copilot seat per user. Long-running advanced work would create new metered consumption.
- New evidence
- Microsoft explicitly places everyday AI inside the USL while Cowork, Code, Autopilot and frontier models consume Copilot Credits—and UBB experiences still require a USL for access.
- Updated view
- Thesis Strengthened, with a narrower boundary: the Consumption Economy is not replacing the Seat Economy. Advanced usage is being added on top. The next proof is whether usage exceeds cannibalization and inference cost.
Platform winners must control the stack from context to cost
- Prior thesis
- Owning a model is insufficient. An enterprise platform must connect context, identity, runtime, workflow, governance and distribution.
- New evidence
- Managed Runtime brings hosting, Microsoft Entra identity, governed data connections, versioning and lifecycle controls into one operating layer. FinOps connects spending policies, credit allocation and usage history to that system.
- Updated view
- The platform-control thesis strengthens. Microsoft can connect distribution, identity, enterprise data, runtime and billing—but this does not yet establish standalone Runtime revenue or higher margins.
03 · EVIDENCE & ANALYSIS
Evidence and analysis
01｜The event was September 25, and product maturity is not uniform
Microsoft announced the new Copilot and FinOps updates on September 25. Cowork is generally available; Code is still rolling through Frontier; Autopilot is in private preview; Managed Runtime is in public preview. The memo separates a confirmed product direction from completed production adoption. An announcement, a preview and scaled revenue are three different levels of evidence.[1][2][6]
02｜Usage billing adds to seats rather than eliminating them
Microsoft divides everyday AI from advanced AI. The USL covers fixed-price capabilities across chat, documents, meetings and analysis. Cowork, Code, Autopilot and some frontier models consume Copilot Credits through UBB—and still require the USL. A better economic model is Users × Subscription + Advanced Work × Credits, not an immediate conversion of every SaaS seat into pure consumption billing.[1][3]
03｜Copilot Credits make agentic work an accountable cost
Copilot Credits are the common metering currency for eligible Microsoft services with usage-based billing. Cowork exposes four broad cost sources—Models, Runtime, Context and Tools—while the admin center can analyze consumption by user, group, agent, service and funding source. Pay-as-you-go is listed at $0.01 per credit, but the credits required for a task vary with the workload, so task cost cannot be inferred from the unit price alone.[4][5][6]
04｜Managed Runtime validates platform control, not standalone revenue
Copilot Managed Runtime supplies Microsoft-hosted execution for apps and code created through Cowork, Code, Copilot Studio and compatible third-party tools, with shared identity, connectors, data access, versioning, health monitoring and policy. It strengthens Microsoft’s position in the enterprise agent control stack. Microsoft has not disclosed a standalone Runtime meter, price or revenue stream, so ‘runtime matters’ cannot be translated directly into ‘runtime is already a separate paid layer.’[2]
05｜FinOps is becoming an adoption gate, not just a budget tool
Long-running agents repeatedly consume models, context, browser activity, tools and runtime resources. Microsoft now lets administrators configure spending policies, user and group access, credit-request routing, threshold notifications and model availability, while users can see balances and usage history. These controls govern access and risk as well as budgets. Without predictable cost, enterprises cannot move comfortably from pilots to frequent production work.[1][4][6]
06｜The investment bridge still runs through revenue, margin and cash
The architecture now answers how advanced work can be billed, where it can run and who can govern it. It does not answer how much customers will consume, how much is net new revenue, or how much inference and support will cost. Microsoft’s distribution, identity and enterprise-data position creates structural leverage; value realization still requires UBB adoption, Azure and Copilot revenue, incremental margin, limited seat cannibalization and free cash flow after capital spending. Product architecture is not financial proof.
04 · INVESTMENT IMPLICATIONS
Industry and asset implications
Microsoft (MSFT)
Subscription plus usage can raise the revenue ceiling per customer; realized value depends on net new consumption, margin and cash conversion.
Enterprise software platforms
Vendors that control context, identity, workflow and governance are better placed to turn agent use into billable, auditable production activity.
Cloud and model supply chain
Longer tasks can raise model and runtime consumption while also increasing inference, storage, tool and support costs. Revenue growth does not automatically improve profit.
Enterprise buyers
Procurement expands from how many seats to which jobs merit metered execution. FinOps and business-outcome measurement become deployment conditions.
This memo discusses public product architecture, business models and listed-company research signals. It is not a recommendation to buy or sell Microsoft or any other security.
05 · VALIDATION & RISKS
What to verify next
Next 30 days | Metering detail
Microsoft publishes credit-consumption rules, model differences, task-cost guidance and spending controls for Code, Autopilot and Managed Runtime.
Failure signal: Capabilities expand while pricing and consumption remain difficult to compare, preventing enterprises from building budgets and ROI models.
Next 30–90 days | Production adoption
Enterprises move Cowork, Code or Autopilot from pilots into repeatable, high-frequency and auditable production workflows, with usage and business outcomes disclosed.
Failure signal: Consumption remains concentrated in trials or low-frequency experiments, and policies mainly restrict use rather than scale valuable work.
Financial loop | Net revenue and margin
UBB and Copilot consumption produce observable net new revenue, with incremental margin covering model, runtime, tool and support costs.
Failure signal: Usage revenue mainly replaces seat revenue, or inference and infrastructure costs absorb the added revenue.
Platform control | Runtime and governance
Managed Runtime expands third-party and enterprise-app adoption, and unified governance supports renewal, expansion and ecosystem distribution.
Failure signal: Enterprises prefer cross-platform runtimes, or governance, permissions and compliance friction slow application deployment.
What would change our view?
The thesis would weaken if advanced-agent usage remains low, UBB mostly cannibalizes existing subscription revenue, or model and runtime costs grow faster than credit revenue. Microsoft’s control-layer advantage would also disappoint if Managed Runtime remains in preview or third-party adoption is limited. Permission errors, data leakage, unpredictable bills, task failures and regulation could all delay production adoption.
06 · FAQ
Key questions
Is Microsoft abandoning seat-based pricing?
No. The USL still covers everyday AI and is the access prerequisite for advanced UBB experiences. The change is that enterprises can pay Copilot Credits for advanced usage in Cowork, Code, Autopilot and frontier models on top of the seat.
What is a Copilot Credit, and what does it cost?
A Copilot Credit is Microsoft’s common metering unit for eligible usage-based AI experiences. Pay-as-you-go is listed at $0.01 per credit; credits consumed by a task vary with models, runtime, context and tool calls.
Is Managed Runtime already a standalone paid product?
That cannot be concluded today. Managed Runtime is in public preview and falls within the expanding cost-management scope, but Microsoft has not published standalone pricing, a meter formula or revenue.
What would prove that the agent economy is producing financial value?
The evidence must link production adoption and repeat task volume to Copilot Credit consumption, net new revenue, incremental gross margin and free cash flow—not product launches or trials alone.
What is the most important risk for the Microsoft thesis?
The central risk is that usage revenue replaces existing seat revenue, runtime cost absorbs the increment, or enterprises delay production deployment because cost, governance and reliability remain difficult to control.
07 · TERMS & SOURCES
Terms, sources and related research
Key terms
- USL
- User Subscription License, the fixed-price Copilot access layer sold per user.
- UBB
- Usage-Based Billing, which charges for actual advanced AI consumption.
- Copilot Credits
- Microsoft’s common metering currency for eligible usage-based AI experiences.
- Managed Runtime
- A Microsoft-hosted environment for operating and governing apps and code inside the Microsoft 365 tenant boundary.
- FinOps for AI
- The discipline of connecting AI budgets, access, consumption and business value.
- Agentic Work
- Multi-step, tool-using or long-running work delegated to an agent.
[1] Microsoft | The new Copilot, UBB and FinOps for AI | 25 September 2026 ↗
[2] Microsoft | Copilot Managed Runtime public preview | 25 September 2026 ↗
[3] Microsoft Learn | Official boundary between the USL and Usage-Based Billing ↗
[4] Microsoft Learn | Copilot Credits and cost management ↗
[5] Microsoft Licensing | Copilot Credit purchase options and published pricing ↗
[6] Microsoft Learn | Cowork governance, access and usage billing ↗
Evidence boundary: product, billing and governance facts come from Microsoft’s 25 September 2026 announcement and official documentation. Microsoft confirms that the USL and UBB coexist, UBB experiences consume Copilot Credits and require a USL, and Managed Runtime is in public preview. Revenue quality, value capture, seat cannibalization and platform control are ACIS interpretations. Microsoft has not disclosed full credit formulas, standalone revenue or unit economics for Code, Autopilot or Managed Runtime. Research and education only; not individualized investment advice.